f2fs: fix 32-bit linking
Subsystems (2)
PATCH 1/1 f2fs: fix 32-bit linking
Not all architectures support get_user() with a 64-bit argument:
ERROR: "__get_user_bad" [fs/f2fs/f2fs.ko] undefined!
Use copy_from_user() here, this will always work.
--- fs/f2fs/file.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
fs/f2fs/file.c +2 -1
@@ -3066,7 +3066,8 @@ static int f2fs_ioc_resize_fs(struct file *filp, unsigned long arg)if (f2fs_readonly (sbi->sb))return -EROFS ; -if (get_user (block_count, (__u64 __user *)arg)) +if (copy_from_user (&block_count, (void __user *)arg, +sizeof (block_count)))return -EFAULT ; ret =f2fs_resize_fs (sbi, block_count);
Signature · 1 line
2.20.0
Quoted 1 line · Arnd Bergmann
Not all architectures support
get_user()with a 64-bit argument:
Which architectures are still missing? I think we finally need to get everyone in line instead of repeatedly working around the lack of minor arch support.
Quoted 7 lines · Christoph Hellwig
Quoted 1 line · Arnd Bergmann
Not all architectures support
get_user()with a 64-bit argument:Which architectures are still missing? I think we finally need to get everyone in line instead of repeatedly working around the lack of minor arch support.
I came across this on arm-nommu (which disablesCONFIG_CPU_SPECTRE) during randconfig testing.
I don't see an easy way to add this in there, short of rewriting the whole __get_user_err() function. Any suggestions?
Arnd
Quoted 5 lines · Arnd Bergmann
I came across this on arm-nommu (which disables
CONFIG_CPU_SPECTRE) during randconfig testing.I don't see an easy way to add this in there, short of rewriting the whole
__get_user_err()function. Any suggestions?
Can't we just fall back to using copy_from_user with a little wrapper that switches based on sizeof()?
Quoted 10 lines · Christoph Hellwig
Quoted 5 lines · Arnd Bergmann
I came across this on arm-nommu (which disables
CONFIG_CPU_SPECTRE) during randconfig testing.I don't see an easy way to add this in there, short of rewriting the whole
__get_user_err()function. Any suggestions?Can't we just fall back to using
copy_from_userwith a little wrapper that switches based onsizeof()?
I came up with something now. It's not pretty, but seems to satisfy the compiler. Not a proper patch yet, but let me know if you find a bug.
This might contain a double uaccess_save_and_enable/uaccess_restore, not sure how much we care about that.
Arnd
index 7e0d2727c6b5..c21cdecadf26 100644
arch/arm/include/asm/uaccess.h +5 -1
@@ -307,6 +307,7 @@ static inline void set_fs(mm_segment_t fs) do { \ unsigned long __gu_addr = (unsigned long)(ptr); \ unsigned long __gu_val; \ + unsigned long long __gu_val8; \ unsigned int __ua_flags; \ __chk_user_ptr(ptr); \ might_fault(); \
@@ -315,10 +316,13 @@ do { \case 1 :__get_user_asm_byte (__gu_val, __gu_addr, err);break ; \case 2 :__get_user_asm_half (__gu_val, __gu_addr, err);break ; \case 4 :__get_user_asm_word (__gu_val, __gu_addr, err);break ; \ +case 8 :__get_user_asm_dword (__gu_val8, __gu_addr, err);break ; \default : (__gu_val) =__get_user_bad (); \ } \uaccess_restore (__ua_flags); \ - (x) = (__typeof__ (*(ptr )))__gu_val; \ + (x) =__builtin_choose_expr (sizeof (*(ptr)) ==8 , \ + (__typeof__ (*(ptr )))__gu_val8, \ + (__typeof__ (*(ptr )))__gu_val); \ }while (0 )
#define __get_user_asm(x, addr, err, instr) \
@@ -373,6 +377,8 @@ do {
\
__get_user_asm(x, addr, err, ldr)#endif
+#define __get_user_asm_dword(x, addr, err) \
+ do { err = raw_copy_from_user(&x, (void __user *)addr, 8) ?-EFAULT : 0; } while (0)
#define __put_user_switch(x, ptr, __err, __fn) \
do { \Russell King - ARM Linux admin
maintainerQuoted 6 lines · Arnd Bergmann
Quoted 5 lines · Christoph Hellwig
Quoted 3 lines · Arnd Bergmann
I came across this on arm-nommu (which disables
CONFIG_CPU_SPECTRE) during randconfig testing.
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
I don't see an easy way to add this in there, short of rewriting the whole
__get_user_err()function. Any suggestions?Can't we just fall back to using
copy_from_userwith a little wrapper that switches based onsizeof()?I came up with something now. It's not pretty, but seems to satisfy the compiler. Not a proper patch yet, but let me know if you find a bug.
Have you checked what the behaviour is when "ptr" is a pointer to a pointer? I think you'll end up with a compiler warning for every case, complaining about casting an unsigned long long to a pointer.
Quoted 45 lines
This might contain a double uaccess_save_and_enable/uaccess_restore, not sure how much we care about that.
Arnd
index
7e0d2727c6b5..c21cdecadf26100644arch/arm/include/asm/uaccess.h +5
-1@@ -307,6 +307,7 @@ static inline void set_fs(mm_segment_t fs) do { \ unsigned long __gu_addr = (unsigned long)(ptr); \ unsigned long __gu_val; \ + unsigned long long __gu_val8; \ unsigned int __ua_flags; \ __chk_user_ptr(ptr); \ might_fault(); \@@ -315,10 +316,13 @@ do { \case 1 :__get_user_asm_byte (__gu_val, __gu_addr, err);break ; \case 2 :__get_user_asm_half (__gu_val, __gu_addr, err);break ; \case 4 :__get_user_asm_word (__gu_val, __gu_addr, err);break ; \ +case 8 :__get_user_asm_dword (__gu_val8, __gu_addr, err);break ; \default : (__gu_val) =__get_user_bad (); \ } \uaccess_restore (__ua_flags); \ - (x) = (__typeof__ (*(ptr )))__gu_val; \ + (x) =__builtin_choose_expr (sizeof (*(ptr)) ==8 , \ + (__typeof__ (*(ptr )))__gu_val8, \ + (__typeof__ (*(ptr )))__gu_val); \ }while (0 )#define
__get_user_asm(x, addr, err, instr)\arch/arm/include/asm/uaccess.h +2
-1@@ -373,6 +377,8 @@ do { \ __get_user_asm(x, addr, err, ldr)#endif +#define __get_user_asm_dword (x, addr, err) \ + do { err = raw_copy_from_user(&x, (void __user *)addr, 8) ? -EFAULT : 0; } while (0)#define __put_user_switch (x, ptr, __err, __fn) \ do { \
Signature · 3 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/ FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up According to speedtest.net: 11.9Mbps down 500kbps up
Quoted 14 lines · Russell King - ARM Linux admin
Quoted 12 lines · Arnd Bergmann
Quoted 10 lines · Christoph Hellwig
Quoted 5 lines · Arnd Bergmann
I came across this on arm-nommu (which disables
CONFIG_CPU_SPECTRE) during randconfig testing.I don't see an easy way to add this in there, short of rewriting the whole
__get_user_err()function. Any suggestions?Can't we just fall back to using
copy_from_userwith a little wrapper that switches based onsizeof()?
Quoted 6 lines
Quoted 2 lines
I came up with something now. It's not pretty, but seems to satisfy the compiler. Not a proper patch yet, but let me know if you find a bug.
Have you checked what the behaviour is when "ptr" is a pointer to a pointer? I think you'll end up with a compiler warning for every case, complaining about casting an unsigned long long to a pointer.
I have built lots of kernels using this patch as a test, though my autobuilder is currently configured to use clang-8, and other compilers or versions might show more warnings.
Quoted 6 lines
arch/arm/include/asm/uaccess.h do { · line 327
Quoted 6 lines
uaccess_restore(__ua_flags); \arch/arm/include/asm/uaccess.h +3
-1- (x) = (__typeof__ (*(ptr )))__gu_val; \ + (x) =__builtin_choose_expr (sizeof (*(ptr)) ==8 , \ + (__typeof__ (*(ptr )))__gu_val8, \ + (__typeof__ (*(ptr )))__gu_val); \ }while (0 )
The __builtin_choose_expr() here is supposed to take care of the case of a pointer type, gcc and clang should both ignore the non-taken branch and only produce warnings for the case they actually use.
Arnd
Quoted 7 lines · Christoph Hellwig
Quoted 1 line · Arnd Bergmann
Not all architectures support
get_user()with a 64-bit argument:Which architectures are still missing? I think we finally need to get everyone in line instead of repeatedly working around the lack of minor arch support.
arch/microblaze/include/asm/uaccess.c does not support get_user() with a size of 8.
Signature · 1 line
~Randy
Jaegeuk Kim
maintainerHi Arnd,
If you don't mind, can I integrate this into the original patch in the queue?
Thanks,
Quoted 28 lines · Arnd Bergmann
Not all architectures support
get_user()with a 64-bit argument:ERROR: "
__get_user_bad" [fs/f2fs/f2fs.ko] undefined!Use
copy_from_user()here, this will always work.Fixes d2ae7494d043 ("f2fs: ioctl for removing a range from F2FS")Signed-off-by Arnd Bergmann <[email protected]>--- fs/f2fs/file.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)fs/f2fs/file.c +2
-1@@ -3066,7 +3066,8 @@ static int f2fs_ioc_resize_fs(struct file *filp, unsigned long arg)if (f2fs_readonly (sbi->sb))return -EROFS ; -if (get_user (block_count, (__u64 __user *)arg)) +if (copy_from_user (&block_count, (void __user *)arg, +sizeof (block_count)))return -EFAULT ; ret =f2fs_resize_fs (sbi, block_count);Signature · 1 line
2.20.0
Quoted 4 lines · Jaegeuk Kim
Hi Arnd,
If you don't mind, can I integrate this into the original patch in the queue?
Yes, I think that would be good anyway, it may take a little longer to fix all the architectures.
Arnd